What is Normal?

The data obtained as a result of the test should be useful to the clinician who orders the test as well as to the subject who generates the result. if a test will not satisfy both of these, then question the need for the test.

We do a diagnostic test for several reasons that include

1. Confirm or establish a diagnosis
2. To judge the severity of a condition
3. As a predictive tool to predict course and prognosis
4. To estimate likely response to therapy and
5. To determine the actual response to therapy

We expect the diagnostic test to then help us classify a person as “normal” or “abnormal”.

Let us look at what “normal” means in the context of a diagnostic test.

The results of an investigation provide you with a normal range- a range of values within which the results are considered normal. In a busy world, we often use these range of normal to categorize some one as normal or abnormal…or to say the result is normal or abnormal.

How do we define normal (in the context of the result of an investigation).

Normal can be defined as “what is normal for the majority of human beings under standard conditions- possibly 95% of the population maybe”.

Let us explore that statement a bit further-it presumes that the distribution of results will follow the NORMAL distribution (the bell curve or the Gaussian distribution) with the extreme 2.5% on either tails (lower and upper) considered as abnormal. Thus, the mean plus/minus two standard deviations is considered as normal. (you can extend that to consider the mean +/- 3SD as normal).

Does this sound reasonable?

Let us check that out.

The normal (Gaussian) distribution extends to infinity on either side, so if the test results do follow the normal distribution, there can be impossibly high hemoglobins or hemoglobins on the wrong side of zero.

Hmm, not biologically plausible? So, maybe then, the results of diagnostic tests do not follow the normal distribution.

Exploring the concept of normal as in normal distribution, if the highest and lowest 2.5% are considered abnormal, it means that 5% of all test results will be considered abnormal. This is interesting because then all diseases will have the same frequency (5%) based on the test results!

That, clearly, is not correct, isn’t it?

So, is normal defined through the normal distribution appropriate?

Don’t forget the population on which the “normal” reference range is built upon- is that a high risk group, a low risk group, a mixed risk group, a group of apparently normal subjects?

Ok, so the normal distribution does not hold very good…why don’t we tweak it a bit to look at the centiles instead.

Thus, any value within the 95th or 90th centile is considered as normal. (You could also say any value less than the 5th or 10th centile is abnormal- example- small for gestational age babies).

We are now working with a limit (0-100) and so have taken care of the extreme values.

How does this hold with the concept of normal? Well, the problem of equal frequency for all diseases persists…

Now, this is the most interesting part of this approach (and the basis for many of the corporate packages that have a lot of investigations- those series of packages that have a multitude of investigations thrown in)…. If we go by the percentile approach, then when we do a test (any test) the probability of that person being called normal is 95%…and the probability of that person being called abnormal is 5%. (remember the 5% is abnormal).

If the person undergoes two independent diagnostic tests (independent in that the tests probe two different things- example, Hemoglobin and say, mammography), the probability of being called normal is now 0.95*0.95=0.90. The likelihood of any patient being called normal is 0.95 raised to the number of tests. Thus, if the package has 20 independent tests (which most of those executive wellness tests have), the probability of being called normal is now ONE IN THREE after the workup! (0.95 raised to the power of 20). The more independent tests you do, the less likely the person will be called normal at the end of the workup!

Now we see that the approach using the 5% (or 10% or 1%) is not entirely appropriate…it has its own limitations that we need to be aware of.

Then how do we determine Normal?

Another approach- commonly used- is the risk factor approach.

Here, the definition of normal is based on prior studies- on precursors or risk factors and on predictors of outcomes…the normal range is now defined as that range that does not cause any additional risk of morbidity or mortality. Example- normal range of diastolic blood pressure (based on cardiovascular risk) may be defined as up to 100 mm of Hg (or 90).

Is that a reasonable approach?

Looks better than the 5% approach…supposedly based on evidence….does not really consider increasing risk with increasing or decreasing levels…(in terms of unit change- example, the risk if greater when diastolic BP increases from 120 to 121 than when it increases from 90-91)..and does not always lead to- altering the risk minimizes risk of outcomes. so intervening to bring back the test result to “normal” may not improve/minimize/change outcomes. (it may also). Then, is the cutoff appropriate?

Another approach- of great interest and often used by the clinician is the diagnostic definition- a range of values beyond which (either above or below) the probability that the target disorder exists is high. otherwise known as the positive predictive value and its various aliases…..this is a much better approach than the 5% and the risk factor approach…but means CLINICIANS HAVE TO KEEP CONSTANT TRACK OF DIAGNOSTIC CUTOFFS AND RANGES

Another common approach is the therapeutic approach- the limit of normal is now set at the point beyond which specific treatments are shown to do more good than harm. Advantage- avoids labeling a person as diseases unless they are going to be treated. Can keep changing so clinicians need to constantly keep pace with developments in therapeutics. Example, the lowering of diastolic bp levels to initiate treatment…..(earlier 130, then 110, 100, now 90?).

Recap- four common approaches to defining normal- the 5% (10% or 1%) and percentile approach, the risk factor approach, the diagnostic definitions and the therapeutic approach…..

The next time you look at the results of an investigation- and the range of normal to determine if the result falls in the “normal range” please do consider how normal is defined!

Isn’t the study of medicine interesting?

How do we determine the clinical usefulness of a diagnostic test?

More Important, How do we apply the result of the test to the individual?

The varying definitions of “Normal” suggest that the result can be a false positive or a false negative…….

Continue to : SENSITIVITY